app.py

import spaces
import gradio as gr
import numpy as np
import random
import torch
from diffusers import DiffusionPipeline

dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=dtype).to(device)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

@spaces.GPU()
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=4, progress=gr.Progress(track_tqdm=True)):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)
    image = pipe(
        prompt=prompt,
        width=width,
        height=height,
        num_inference_steps=num_inference_steps,
        generator=generator,
        guidance_scale=0.0
    ).images[0]
    return image, seed

# Example prompt
example_prompt = "A vibrant red origami crane on a white background, intricate paper folds, studio lighting"

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# FLUX.1 [schnell] Image Generator")
    
    with gr.Row():
        with gr.Column(scale=2):
            gr.Markdown("""
            ## About FLUX.1 [schnell]
            - Fast text-to-image model optimized for local development and personal use
            - Part of the FLUX.1 model family by Black Forest Labs
            - Open-source: Available under Apache 2.0 license
            - Supports resolutions between 0.1 and 2.0 megapixels
            - Outperforms many larger models in quality and prompt adherence
            - Uses advanced transformer architecture with flow matching techniques
            - Capable of generating high-quality images in just a few inference steps
            """)
        
        with gr.Column(scale=3):
            prompt = gr.Textbox(label="Prompt", placeholder="Enter your image description here...", value=example_prompt)
            run_button = gr.Button("Generate")
            result = gr.Image(label="Generated Image")
    
    gr.Markdown("""
    ## Example Prompt
    Try this example prompt or modify it to see how FLUX.1 [schnell] performs:
    ```
    A vibrant red origami crane on a white background, intricate paper folds, studio lighting
    ```
    """)
    
    with gr.Accordion("Advanced Settings", open=False):
        seed = gr.Slider(minimum=0, maximum=MAX_SEED, step=1, label="Seed", randomize=True)
        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
        width = gr.Slider(minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024, label="Width")
        height = gr.Slider(minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024, label="Height")
        num_inference_steps = gr.Slider(minimum=1, maximum=50, step=1, value=4, label="Number of inference steps")
        
        gr.Markdown("""
        **Note:** FLUX.1 [schnell] is optimized for speed and can produce high-quality results with just a few inference steps.
        Adjust the number of steps based on your speed/quality preference. More steps may improve quality but will increase generation time.
        """)
    
    gr.Markdown("""
    ## Additional Information
    - FLUX.1 [schnell] is based on a hybrid architecture of multimodal and parallel diffusion transformer blocks
    - It supports various aspect ratios within the 0.1 to 2.0 megapixel range
    - The model uses bfloat16 precision for efficient computation
    - For optimal performance, running on a CUDA-enabled GPU is recommended
    - For more details and other FLUX.1 variants, visit [Black Forest Labs](https://blackforestlabs.ai)
    """)
    
    run_button.click(
        infer,
        inputs=[prompt, seed, randomize_seed, width, height, num_inference_steps],
        outputs=[result, seed]
    )

demo.launch()